HCV genome-wide genetic analyses in context of disease progression and hepatocellular carcinoma

Transcription

1 Washington University School of Medicine Digital Open Access Publications 2014 HCV genome-wide genetic analyses in context of disease progression and hepatocellular carcinoma Maureen J. Donlin Saint Louis University Elena Lomonosova Saint Louis University Alexi Kiss Washington University School of Medicine in St. Louis Xiaohong Cheng Saint Louis University Feng Cao Saint Louis University See next page for additional authors Follow this and additional works at: Recommended Citation Donlin, Maureen J.; Lomonosova, Elena; Kiss, Alexi; Cheng, Xiaohong; Cao, Feng; Curto, Teresa M.; Di Bisceglie, Adrian; and Tavis, John E.,,"HCV genome-wide genetic analyses in context of disease progression and hepatocellular carcinoma." PLoS One.9,7. e (2014). This Open Access Publication is brought to you for free and open access by Digital It has been accepted for inclusion in Open Access Publications by an authorized administrator of Digital For more information, please contact

2 Authors Maureen J. Donlin, Elena Lomonosova, Alexi Kiss, Xiaohong Cheng, Feng Cao, Teresa M. Curto, Adrian Di Bisceglie, and John E. Tavis This open access publication is available at Digital

3 HCV Genome-Wide Genetic Analyses in Context of Disease Progression and Hepatocellular Carcinoma Maureen J. Donlin 1,2, Elena Lomonosova 1, Alexi Kiss 1 a, Xiaohong Cheng 1, Feng Cao 1,2 b, Teresa M. Curto 5, Adrian Di Bisceglie 1,4, John E. Tavis 1,3 * 1 Department of Molecular Microbiology and Immunology, Saint Louis University School of Medicine, Saint Louis, Missouri, United States of America, 2 Edward A. Doisy Department of Biochemistry and Molecular Biology, Saint Louis University School of Medicine, Saint Louis, Missouri, United States of America, 3 Saint Louis Liver Center, Saint Louis University School of Medicine, Saint Louis, Missouri, United States of America, 4 Department of Internal Medicine, Saint Louis University School of Medicine, Saint Louis, Missouri, United States of America, 5 New England Research Institutes, Inc., Watertown, Massachusetts, United States of America Abstract Hepatitis C virus (HCV) is a major cause of hepatitis and hepatocellular carcinoma (HCC) world-wide. Most HCV patients have relatively stable disease, but approximately 25% have progressive disease that often terminates in liver failure or HCC. HCV is highly variable genetically, with seven genotypes and multiple subtypes per genotype. This variation affects HCV s sensitivity to antiviral therapy and has been implicated to contribute to differences in disease. We sequenced the complete viral coding capacity for 107 HCV genotype 1 isolates to determine whether genetic variation between independent HCV isolates is associated with the rate of disease progression or development of HCC. Consensus sequences were determined by sequencing RT-PCR products from serum or plasma. Positions of amino acid conservation, amino acid diversity patterns, selection pressures, and genome-wide patterns of amino acid covariance were assessed in context of the clinical phenotypes. A few positions were found where the amino acid distributions or degree of positive selection differed between in the HCC and cirrhotic sequences. All other assessments of viral genetic variation and HCC failed to yield significant associations. Sequences from patients with slow disease progression were under a greater degree of positive selection than sequences from rapid progressors, but all other analyses comparing HCV from rapid and slow disease progressors were statistically insignificant. The failure to observe distinct sequence differences associated with disease progression or HCC employing methods that previously revealed strong associations with the outcome of interferon a- based therapy implies that variable ability of HCV to modulate interferon responses is not a dominant cause for differential pathology among HCV patients. This lack of significant associations also implies that host and/or environmental factors are the major causes of differential disease presentation in HCV patients. Citation: Donlin MJ, Lomonosova E, Kiss A, Cheng X, Cao F, et al. (2014) HCV Genome-Wide Genetic Analyses in Context of Disease Progression and Hepatocellular Carcinoma. PLoS ONE 9(7): e doi: /journal.pone Editor: Naglaa H. Shoukry, University of Montreal Hospital Research Center (CRCHUM), Canada Received April 9, 2014; Accepted July 1, 2014; Published July 31, 2014 This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. Data Availability: The authors confirm that all data underlying the findings are fully available without restriction. The HCV sequence data are currently available from Genbank numbers for HCV sequences from the HCC cohort are: KC KC (HCC) and KC KC (cirrhotic controls). Genbank numbers for HCV sequences from the HALT-C patients are: JX JX (time point 1, rapid progressors); JX JX (time point 1, slow progressors); JX JX (time point 2, rapid progressors), and JX JX (time point 2, slow progressors). All other relevant data are within the paper and its Supporting Information files. Funding: Funding provided by National Institute of Allergy and Infectious Disease grant number DK to JT and National Cancer Institute grant number CA to JT. The authors acknowledge the HALT-C Trial group for providing samples and data for this publication. The HALT-C Trial was funded through contracts from the National Institute of Diabetes & Digestive & Kidney Diseases. Additional support was provided by the National Institute of Allergy and Infectious Diseases, the National Cancer Institute, the National Center for Minority Health and Health Disparities and by General Clinical Research Center and Clinical and Translational Science Center grants from the National Center for Research Resources, National Institutes of Health. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Center for Research Resources or the National Institutes of Health. Additional funding to conduct this study was supplied by Hoffmann-La Roche, Ltd., through a Cooperative Research and Development Agreement with the National Institutes of Health. The HALT-C Trial was registered with clinicaltrials.gov (#NCT ). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing Interests: Both MD and JT are Academic Editors for PLOS ONE. The authors both confirm that this does not alter their adherence to PLOS ONE Editorial policies and criteria. Additionally, no funding was directly received from Hoffmann-La Roche in support of this project. The HALT-C clinical trial from which some of the samples were derived received funding from Hoffman La Roche. However, HALT-C was funded and conducted independently of the authors viral genetics study, so no commercial funding was directly spent in support of this viral genetics project. Effort expended by the HALT-C personnel in support of this project was paid for by National Institutes of Health grant DK to JT through a subcontract with the HALT-C Data Coordinating Center at the New England Research Institute. Despite this separation in funding sources, the authors included the funding sources for the HALT-C trial in our disclosures to provide full transparency. Therefore, the disclosure in the initial submission was complete and accurate, and an amendment of the conflict of interest statement does not appear to be needed. The authors attest that the funding of the parental HALT-C study by a commercial sources does not alter their adherence to all PLOS ONE policies on sharing data and materials. * tavisje@slu.edu a Current address: Division of Dermatology, Washington University in St. Louis School of Medicine, St. Louis, Missouri, United States of America b Current address: John Cochran Division, Department of Veteran s Affairs Medical Center, St. Louis, Missouri, United States of America PLOS ONE 1 July 2014 Volume 9 Issue 7 e103748

4 Introduction Hepatitis C virus (HCV) is a Hepacivirus that infects hepatocytes and some lymphocytes [1,2]. It chronically infects about million people world-wide, resulting in about 350,000 deaths annually [3,4]. Disease caused by HCV ranges from asymptomatic infection to severe hepatitis, with most people having some degree of ongoing liver damage [1,5]. Roughly 25% of chronically infected individuals have progressive disease, where liver pathology proceeds from hepatitis of gradually worsening severity, to hepatic fibrosis, cirrhosis, and often to fatal liver failure or hepatocellular carcinoma (HCC). The rate of progression along this spectrum varies from a few years in exceptionally rapid progressors to many decades in slow progressors, with relatively slow disease progression being the norm. HCV-induced liver disease is primarily caused by hepatic inflammation and anti-hcv immune responses [6 8]. Direct cytopathic effects from viral replication may contribute to disease, but they are believed to be secondary to immune-mediated damage. HCV s,9,600 nucleotide positive-polarity RNA genome encodes a polyprotein of,3100 amino acids that is cleaved into 10 mature proteins (Fig. 1). The genome is surrounded by a capsid composed of the viral core protein, and the capsid is enclosed by a lipid envelope containing the viral glycoproteins E1 and E2. The non-structural proteins (P7-NS5B) replicate the viral RNA, and virions are secreted from the cell non-cytolytically [9,10]. The HCV genome is highly variable, with seven genotypes that are less than 72% identical at the nucleotide level [11]. Within the genotypes, subtypes with nucleotide identities of 75 86% may occur. Individual isolates of a given subtype are typically,92 96% identical, and as HCV replicates as quasispecies, multiple variants differing by up to a few percent exist within individual patients. The viral 59 untranslated region, the core gene, and the extreme 39 end of the genome are relatively well conserved, and two hypervariable regions within the envelope proteins, the 39 end of the NS5A gene, and parts of the 39 untranslated region are very poorly conserved. Until recently the standard treatment for chronic HCV infection was pegylated interferon a (IFNa) plus ribavirin for 24 to 48 weeks, which resulted in clearance of the virus [sustained viral response (SVR)] in 50 60% of genotype 1 patients [12,13]. In 2011, two inhibitors of the HCV NS3 protease, telaprevir and boceprevir, were approved for use in conjunction with interferon a in HCV genotype 1 patients that improved SVR rates to,75% [14,15]. A third inhibitor of the NS3 protease (simeprevir) and a nucleoside analog that targets the NS5B RNA polymerase (sofosbuvir) were approved in 2013 [16 18], increasing efficacy of the triple-therapy combinations. However, stimulation of the interferon response remains key to efficacy of the existing triple therapies, and HCV treatment will remain dependent on interferon a until sets of direct-acting drugs with sufficient efficacy to eradicate the virus by themselves is approved, as is expected to happen [19]. HCV s genetic variation has a major impact on success of both interferon a-based therapy and direct inhibitor-based treatments. Telaprevir and boceprevir are approved exclusively for patients infected with HCV genotype 1 [20], whereas simeprevir is approved for use against both genotype 1 and 4 infections [16]. Most experimental direct-acting agents are also genotype-specific [21,22]. Interferon plus ribavirin therapy clears genotype 1 infections much less well than genotype 2 and 3 infections (,50% compared to.80%, respectively) [23,24]. We previously found that high genetic variation in the consensus sequences of the HCV core, NS3, and NS5A genes was tightly correlated with failure of interferon a plus ribavirin therapy [25 27]. Importantly, all three of these genes can counteract the type 1 interferon response [28]. We interpreted this association to indicate that high viral variability impairs the ability of HCV s interferon-suppressive proteins to counteract the heightened type 1 interferon responses induced by therapy. We and others also found that,10% of HCV s,3000 amino acid positions covary with at least one other position, and that these covariances link together into a genomewide network of covarying positions [27,29]. These networks are different among HCV sequences from responders and nonresponders to interferon plus ribavirin therapy [27,30,31], implying a coordinated role for sequence variation throughout the viral genome in antagonizing interferon responses. The impact of HCV genetic variation on viral pathology is less clear. It is well accepted that genotype 3 causes steatosis more frequently than the other HCV genotypes [32,33]. Furthermore, HCV infection can elevate levels of the pro-inflammatory cytokine IL8 [34] through activation of the IL8 promoter by core, NS4B, and/or NS5A [35 37], and there is a direct correlation between core sequence variation, ALT levels, and IL8 promoter activation [38]. Other associations between HCV genetic variation and pathology are less well accepted. Some studies found little evidence for virulence differences between the major HCV genotypes [39,40], but most studies found differences, such as genotype 1 being more virulent than genotype 2 [41,42]. Most studies have found genotype 1b to be more virulent and more highly associated with HCC than other genotypes [43 47]. However, these associations have not been apparent in other studies [48,49], and some of the higher virulence of 1b has been suggested to be due to accidental selection bias in the patient populations [50]. Stronger evidence exists for a role of HCV genetic variation on HCC development. Most genetic analyses of HCV in the context of HCC have focused on the HCV core and NS5A genes. The HCV core protein has been reported to promote cellular transformation in tissue culture [51] and in some animal models [52,53], and several studies have found an association between variations in the core coding sequence and the likelihood of Figure 1. HCV genome. The HCV genome contains 59 and 39 untranslated regions and a single, long open reading frame that encodes 10 proteins. The mature viral proteins encoded within the open reading frame and their major functions are indicated. Reprinted from [25] under the creative commons license. doi: /journal.pone g001 PLOS ONE 2 July 2014 Volume 9 Issue 7 e103748

5 developing HCC [54 58]. Akuta et. al identified two core amino acid positions (70 and 91) where non-wild type residues were significantly associated with HCC in genotype 1b patients [55 57]. Furthermore, Fishman et. al. examined core nucleotide positions and their putative effect on known RNA structures in subtype 1b and identified several positions where substitutions were associated with increased risk of HCC [54]. Inhibition of PKR activity by the NS5A PKR binding site has been shown to be needed for cellular transformation and tumorigenicity in nude mice [59], but both low [60,61] and high [62] diversity of the PKR binding site have been associated with HCC. Studies in which HCV variation has been described at just the subtype level found that 1b is associated with a higher risk of HCC than 1a [46,47,63]. Three studies examined full-length HCV genomes at the sequence level in the context of HCC [64,65] [66], and each study identified a small number of amino acid positions where genetic variation was significantly associated with HCC. We hypothesized that HCV genetic variation may be associated with differential virulence in HCV, specifically with the rate of advancement of liver disease and/or development of HCC. This hypothesis was based on our identification of clear HCV genetic patterns that were associated with outcome of interferon-based antiviral therapy [25 27]. The null hypothesis was that environmental and/or host-specific factors were dominant in determining the differential disease outcomes. Two independent patient sets were employed to assess this hypothesis. The first set was used to evaluate association of HCV genetic variation with development of HCC. These patients were identified through the Liver Cancer Research Network (LCRN; [67]). The second set was used to assess the role of HCV genetic diversity in the rate of disease progression. These patients were derived from the untreated observational control arm of the HALT-C clinical trial, which was a multi-center, randomized controlled study designed to determine if long-term interferon a treatment would ameliorate HCV s pathology [68]. Our strategy was to determine the consensus sequence for the full HCV coding region by direct sequencing of nested reverse-transcription-pcr products, and then to compare HCV genetic patterns in HCC vs. non-hcc patients for the cancer cohort or the slow vs. rapid progressors from the HALT-C cohort. Methods Ethics Statement This study was approved by the Saint Louis University Biomedical Institutional Review Board (HCC cohort, IRB#1570; HALT-C cohort IRB#14138). All participants provided written informed consent to participate in the parent HALT- C and LCRN studies; this informed consent included granting permission for use of de-identified samples for study-approved ancillary studies such as this. This informed consent procedure was approved by the IRBs for the parental study, and each patient s informed was documented and filed by the parental studies. Sequencing the HCV open reading frame HCV RNAs were isolated from patient serum and cdnas were synthesized as previously described [69,70]. For the HCC samples and cirrhotic controls, cdnas were sequenced with the nested reverse transcriptase-pcr and direct sequencing methods we previously employed for Virahep-C samples [26,69]. cdnas from the HALT-C samples were sent to the Broad Institute for nested reverse transcriptase-pcr and direct sequencing of the overlapping amplicons by the chain-termination method as described [70]. Approximately 50% of the viral sequence data were obtained by this approach. The remaining data for the HALT-C sequences were obtained employing our higher-sensitivity nested reverse transcriptase-pcr and direct sequencing methods [26,69]. The extreme 39 end of the HCV open reading frame could not be obtained for all patients. Consequently, the sequences were truncated at aa 8991 for the cancer cohort and at aa 8994 for the HALT-C patients to ensure equal coverage of all genomic regions in the analyses. This eliminated the 14 C-terminal codons for the cancer cohort and the 13 C-terminal codons for the HALT-C sequences. Genbank numbers for HCV sequences from the HCC cohort are: KC KC (HCC) and KC KC (cirrhotic controls). Genbank numbers for HCV sequences from the HALT-C patients are: JX JX (time point 1, rapid progressors); JX JX (time point 1, slow progressors); JX JX (time point 2, rapid progressors), and JX JX (time point 2, slow progressors). The list of sequence IDs, accession numbers and experimental groups for both patient cohorts are in Table S1. Clonal sequencing in the E2 gene Twelve clones encompassing the amino-terminal region of the E2 glycoprotein (aa in the HCV polyprotein) that included the hypervariable region 1 (aa ) from each of six HCC and six cirrhotic control patients were cloned for quasispecies analyses. HCV RNAs were isolated and cdna was synthesized as was done for the direct sequencing. HCV sequences were amplified by nested PCR from the cdnas under high-fidelity conditions employing the Hotstart HiFidelity Polymerase kit (Qiagen). The PCR products were cloned and independent clones were randomly selected for sequencing. Sequence analyses All analyses except dinucleotide frequency analyses and codon selection biases were conducted at the amino acid level. Consensus population-wide reference sequences were derived from 107 fulllength genotype 1b or 103 genotype 1a ORFs downloaded from Genbank in January, Sequence alignments were done with Muscle [71]. Positions that varied relative to the genotype 1a or 1b reference consensus sequences were identified with the EMBOSS program Infoalign 4501 [72]. Mean genetic distance was calculated using the p-distance algorithm in the MEGA v. 5 DNA analysis package [73]. The codon selection analysis based on the ratio of dn/ds substitutions was done using the single likelihood ancestor counting (SLAC) method with the HKY85 substitution mode and a significant level of p,0.05 [74]. The predicted frequency of specific dinucleotide pairs within an ORF was calculated by multiplying the frequency in the ORF of both bases in the pair by the length of the ORF using customized PERL scripts. The observed base and dinucleotide compositions were counted directly using customized PERL scripts. Amino acid covariance analyses All possible amino acid covariances within the HCV open reading frame were determined employing the observed-minus expected-squared algorithm with a 1% false discovery rate as we have previously described [27,75]. Networks in the covariance data were graphed employing Cytoscape [76]. Network metrics were calculated employing the Network Analyzer plug-in for Cytoscape [77]. Statistical analyses Positions of skewed amino acid variance between the groups of sequences were identified by comparing positions of variance in PLOS ONE 3 July 2014 Volume 9 Issue 7 e103748

6 each group with a Mann-Whitney ranked sums test. Differences in the average protein distances, the average number of variations/ sequence and dinucleotide frequencies were compared with a t- test. Statistical analyses were carried out using SPSS v19 (IBM Corporation, Armonk, NY). Baseline variables in the rapid and slow progressor groups were compared using the chi-square test, the t-test, or the Wilcoxon rank-sum test using SAS v.9.3 (SAS Institute, Cary, NC). Results Patient selection and sequencing to evaluate association of HCV genetic variation with HCC Fifty patients were identified through the Liver Cancer Research Network (LCRN) and Dr. Di Bisceglie s practice at Saint Louis University for the cancer cohort. All patients were infected with HCV subtype 1b and had a clinical diagnosis of cirrhosis at or prior to sample collection. Exclusion criteria included co-infection with HBV or HIV, evidence of alcohol abuse, and evidence of other liver diseases including non-alcoholic fatty liver disease or hemochromatosis. HCC patients had a definite or presumed HCC diagnosis at sample collection. Definite HCC was biopsy-proven HCC or the presence of a new defect within the liver noted on imaging studies with a serum alphafetoprotein (AFP) level of.1,000 ng/ml. Presumed HCC was three separate imaging techniques suggestive of HCC, a new hepatic defect followed by massive hepatic involvement and death, or a new hepatic defect with increasing size or increasing serum AFP. Cirrhotic controls were cirrhotic (confirmed by liver biopsy, with Metavir score $4) but had no clinical evidence of HCC at the time of sample collection. HCC was excluded in the controls by routine ultrasound surveillance every 6 to 12 months according to the AASLD practice guideline on management of HCC. The HCC and cirrhotic control patient groups were matched by age and sex. The annual incidence rate of HCC in cirrhotic HCV-infected patients is 1 to 4% [78]. Therefore, our power calculations assumed that two of the 25 cirrhotic controls (8%) that were cancer-free at sample collection would develop HCC within a few years. Using 25 controls yielded.80% power at a = 0.05 to detect genetic differences similar to what we had observed with the Virahep-C samples between the HCC and cirrhotic groups, even with this high degree of contamination of the controls. Consensus sequences for the full HCV coding region were obtained from serum-derived RNA employing the nested reverse transcriptase-pcr and direct sequencing methods we previously employed [26,69]. We were unable to sequence the full coding region from three HCC patients so these sequences were excluded. The HCC and cirrhotic control groups remained statistically indistinguishable for age and sex following exclusion of these three patients (Table 1). HCV positional sequence differences associated with HCC To identify amino acid positions in the HCV sequence that differed consistently between the HCC and cirrhotic control sequences, we aligned the sequences and examined amino acid distributions at all 2997 positions. The amino acid distributions at 25 aa positions were significantly different between the HCC and cirrhotic control samples, with p-values ranging from to (Table 2). As a control to determine the frequency of chance associations in this analysis, the 47 sequences were randomly resorted into five sets of 25 and 22 sequences and positions where the amino acid distribution differed significantly between these pairs of biologically irrelevant groups were identified. We observed a mean of 15.2 (10 22) positions that differed with a mean p-value of (0.001 to 0.049) in these control comparisons. The larger number of significantly different positions in the HCC versus cirrhotic case compared to the scrambled control sequence sets suggests some of the 25 positions of skewed variance between HCC and cirrhotic controls may be associated with a biological difference between the two groups. Four of the positions that were significantly associated with HCC occurred in the very small (63 residue) p7 gene. HCV consensus sequence diversity differences are not associated with HCC To determine if there were diversity differences between the HCC and cirrhotic control sequences, pairwise genetic distances between all samples within the HCC and cirrhotic controls groups were calculated, and then the average pairwise distances were compared between the two groups. The mean pairwise differences for the HCC and cirrhotic control groups (0.081 vs , respectively) were not significantly different. A more sensitive method to measure genetic diversity is to quantify the number of variations relative to a population-wide consensus reference sequence for each sample. Therefore, each sample was aligned to a subtype 1b population-wide reference sequence, and the number and identity of variations were recorded for each sample as we have done before [25,26,79]. No significant differences were found between the HCC and cirrhotic controls for either total number of variations in the two groups or the number of variations that were unique to either the HCC or cirrhotic groups. This held true when the entire polyprotein was evaluated as a single unit and when the viral genes were considered individually. HCV quasispecies patterns in the E2 HVR region are not strongly associated with HCC To determine if there were genetic differences between the HCC and cirrhotic groups at the quasispecies level, we sequenced 12 independent clones covering the 27 amino acid-long E2 hypervariable region (HVR) plus 66 amino acids downstream of the HVR from each of six randomly-selected patients in both the Table 1. Age and gender of patients from whom the HCC and cirrhotic samples were derived. Cirrhotic control HCC P value 1 Number of patients Gender (M/F) 19/6 16/6 ns Age (mean 6 SD) ns Age (range) ns 1 ns, non-significant. doi: /journal.pone t001 PLOS ONE 4 July 2014 Volume 9 Issue 7 e103748

7 Table 2. Positions where the distribution of amino acids differ in HCC and cirrhotic control sequences. Position Protein P value 75 Core E E E E E E E E p p p p NS NS NS NS NS NS4A NS5A NS5A NS5A NS5A NS5B NS5B doi: /journal.pone t002 HCC and cirrhotic groups. The number of amino acid differences relative to a genotype 1b population reference per patient was not significantly different between sequences from the HCC and cirrhotic samples. Amino acid pairwise distances were determined within the set of 12 sequences for each patient as a measure of the quasispecies diversity. The mean pairwise protein genetic differences were slightly higher in the E2 region (0.066 vs ) and the HVR region (0.278 vs ) for the HCC samples compared to the cirrhotics. Sequence complexity within the 12 sequences per patient was also assessed. The HCC patients had an average of 9.2 unique E2 sequences per patient compared to 7.2 for the cirrhotic controls, and the HCC samples had an average of 7.5 unique HVR sequences per patient compared to 6.5 for the cirrhotics. Similar results were obtained when the data were analyzed at the nucleotide level. Thus, the HCV sequences in the HCC patients appeared to be slightly more diverse and complex than in the cirrhotic controls, but these differences were not statistically significant. There was no evidence of positive selection in these sequences. Overall, no prominent differences in the quasispecies spectra in the HCC and control patients were detected. Selective pressures associated with HCC We examined the HCC and cirrhotic control sequences for differences in selective pressure at all 2997 codons using the SLAC method with the HKY85 substitution mode in order to identify the codons under positive or negative selection [74]. 825 of the 2997 codons were under negative selection and 12 codons were under positive selection in the HCC sequences, while 900 codons were under negative selection and 13 codons under positive selection in the cirrhotic controls (Table 3). Only three of the positivelyselected codons were shared between the two groups. To help evaluate whether these selective differences may be associated with disease state or may simply represent selective pressures on the HCV population as a whole, we randomly sampled six sets of 22 or 25 HCV 1b coding sequences of the same length (2997 codons) from Genbank and examined them for codon selection differences. Six of the 25 positions under positive selection in the HCC or cirrhotic control sequences were not under positive selection in any of the six randomly selected sequence sets (bold in Table 3). This indicates that the positive selection pressures on most of the sites we identified were probably unrelated to the patient s disease state, but that selection at the six codons unique to the HCC or cirrhotic patients may reflect evolutionary pressures associated with these advanced disease states. UU and UA dinucleotide frequency differences are not associated with HCC RNase L is an endoribonuclease that cleaves RNA at singlestranded UA and UU dinucleotides [80]. RNAse L contributes to the innate immune responses against many viruses. We and others have shown that RNase L exerts evolutionary pressure on HCV genomes, as evidence by a reduced frequency of UU and UA dinucleotides than would be expected by chance [26,81]. We extended these analyses by determining the ratio of observed/ predicted dinucleotide frequencies for every possible dinucleotide pair for each of the HCV sequences from the HCC and cirrhotic patients. All of the samples showed the predicted reduced frequency of UA and UU dinucleotide pairs, with an average observed/expected ratio of 0.81 and 0.94 respectively. However, there were no significant differences in the frequency of any dinucleotide pair between HCC and the cirrhotic controls. Amino acid covariance patterns associated with HCC Next, we asked whether differences in genome-wide amino acid covariance networks distinguished the HCC and cirrhotic control sequences. This analysis was based on our previous detection of prominent differences in the networks from responders and nonresponders to interferon-based therapy [27,31]. Amino acid covariance networks were generated for the HCC and cirrhotic controls as previously described [75]. As has been observed for other HCV sequence sets [27,29,31,75], covariance networks containing residue positions from all 10 proteins that had a huband-spoke topology were observed. However, the HCC network had many fewer nodes and was much less tightly connected than the cirrhotic network (Table 4). The less-connected nature of the HCC network was obvious visually, as it formed two major and many smaller networks instead of a single large network as was formed by the cirrhotic sequences (Fig. 2). We previously found that detection of covariances was sensitive to the number of sequences employed, with sequences being on the lower end of the useful range [75]. Therefore we asked if this network integrity difference was due to the fewer number of sequences in the HCC set by randomly selecting six sets of 22 cirrhotic sequences and generating analogous amino acid covariance networks. All of the networks contained very similar numbers of nodes and covarying pairs as were found in the network with 25 cirrhotic sequences, suggesting that the network is robust to the loss of a few sequences (Table 4). To test the possibility that the network connectivity differences were due to a PLOS ONE 5 July 2014 Volume 9 Issue 7 e103748

8 Table 3. Codons under positive selection in the HCC and cirrhotic control sequences. Codon Gene Group dn-ds P value 75 Core Cirrhotic E2 Cirrhotic E2 Cirrhotic E2 Cirrhotic E2 Cirrhotic E2 Cirrhotic E2 Cirrhotic E2 Cirrhotic E2 Cirrhotic NS3 Cirrhotic NS5A Cirrhotic NS5B Cirrhotic NS5B Cirrhotic E2 HCC E2 HCC E2 HCC E2 HCC E2 HCC E2 HCC NS2 HCC NS2 HCC NS3 HCC NS5B HCC NS5B HCC Bold indicates positions unique to the HCC or cirrhotic sequence alignments compared to alignments of randomly selected HCV control sequences. doi: /journal.pone t003 random sampling of HCC sequences, we generated analogous covariance networks from two other sets of full-length HCV sequences from HCC-positive patients [64,65] (15 or 13 sequences). We also combined these two sequence sets and generated two networks from randomly selected sets of 22 sequences from the combined set of 28 sequences; Table 4). In all 4 of these HCV covariance networks derived from HCC patients outside of our patient cohort, the covariances formed a single, highly connected network with network parameters similar to the cirrhotic network and to the previously published HCV networks [27,75]. This suggests that the fragmented covariance network observed with our HCC sequences is unlikely to be a general feature associated with HCC patients. Positional differences in the HCV core gene associated with HCC Genetic variations at 11 nucleotide [54] and two amino acid [55 57] positions in the core gene have been associated with HCC. Therefore, we evaluated these genetic signatures in our sequences. The HCC and the cirrhotic control sequences both predominantly carried the control-type rather than the HCC-type sequences at 8 of the 11 sites that were associated with HCC by Fishman et al. [54] (Table 5). Furthermore, the distribution of HCC- and control-type sequences at all of these positions was quite similar in our HCC and cirrhotic sequence sets. Sequence data for four of these positions are available for an independent set of cirrhotic patients [62]. The sequence patterns in this independent cirrhotic cohort were nearly identical to the patterns we observed at all 4 positions (Table 5). Very similar results were obtained when we analyzed amino acid variation at core residues 70 and 91 (these codons include nucleotides 209 and 271, respectively). Having a residue other than arginine at position 70 or leucine at position 91 has been associated with HCC [55 57]. The majority of our HCC and cirrhotic sequences had the cancer signature at both positions 70 and 91 (Table 5). These results were corroborated by the external set of cirrhotic sequences. Therefore, the sequence patterns at all 11 nucleotide positions and both of the amino acid positions in core that have been previously associated with HCC were distributed almost identically among the HCC and cirrhotic sequences, with non-cancer sequence patterns predominating at 8 of the 11 nucleotide positions. Patient selection and sequencing to evaluate association of HCV genetic variation with rate of disease progression The HALT-C trial evaluated the efficacy of long-term low-dose interferon a therapy on the rate of progression of liver disease in HCV patients who had previously failed interferon plus ribavirin therapy [68]. The study included a large observational control arm that did receive long-term interferon therapy, and hence provides a unique resource for studying HCV s role in disease progression. Sixty patients from the observational arm of the HALT-C study who had been followed for four years were therefore identified for analysis; 30 were slow progressors and 30 were rapid PLOS ONE 6 July 2014 Volume 9 Issue 7 e103748

9 Table 4. Basic amino acid covariance network parameters for the HCC and cirrhotic sequences. Network Number of sequences Number of nodes Number of edges Average number of neighbors Network density Experimental sequences HCC Cirrhotic Randomly sampled sets of 22 cirrhotic sequences Set Set Set Set Set Set External HCC sequence sets Pooled set Pooled set Takahashi et al Nagayama et al Sequences obtained from [65]. 2 Sequences obtained from [64]. doi: /journal.pone t004 progressors. All patients were infected with HCV subtype 1a, had failed prior interferon a plus ribavirin therapy, and had Ishak fibrosis scores at entry to HALT-C of 3 or 4. Patients co-infected with HBV or HIV were excluded. Patients were defined as Rapid Progressors if any of the standard HALT-C outcome criteria were met during the observation period: having a Child-Turcotte- Pugh (CTP) score $7 on two consecutive study visits, variceal hemorrhage, ascites, bacterial peritonitis, encephalopathy, advancement of the Ishak fibrosis score $2 points compared to the initial score, development of HCC, or dying from liver-related causes. Slow Progressors were defined as patients who did not meet any of these HALT-C outcomes during the observational Figure 2. Amino acid covariance networks for the HCC and cirrhotic sequences. Amino acid covariances within alignments of the HCV cirrhotic (left) and HCC (right) sequences were graphed with the covarying positions (nodes) represented as circles and the covariances between the positions (edges) as lines. The size of the nodes is proportional to the number of edges that they contact. Yellow nodes are within structural proteins and green nodes are in non-structural proteins. The amino acid residue position numbered relative to the HCV polyprotein is indicated in the larger nodes. doi: /journal.pone g002 PLOS ONE 7 July 2014 Volume 9 Issue 7 e103748

10 Table 5. Positions in core associated with HCC. Residue identity Control-type/HCC-type residues (number of sequences) Residue number 1 Control-type HCC-type Cirrhotic sequences HCC sequences External cirrhotic sequences 2 Nucleotide positions 3 36 A G/C 25/0 21/1 22/5 78 U C 1/24 2/20 1/ G A 7/18 5/17 6/ U/C A 3/22 5/17 6/ U C/A 21/4 17/5 na C U 25/0 22/0 na 408 C U 24/1 21/1 na 435 G A/C 25/0 22/0 na 465 C U 24/1 20/2 na 481 G A 25/0 21/1 na 546 G A/C 23/2 19/3 na Amino acid positions 5 70 R non-r 7/18 5/17 5/21 91 L M 3/22 5/17 6/19 1 Numbered relative to the start of the polyprotein. 2 Sequences obtained from [62]. 3 Identified in [54], genotype 1b. 4 na, not available. 5 Identified in [55,57]. doi: /journal.pone t005 period. The slow responders included 23 patients whose HCV titers never became undetectable during failed interferon-based antiviral therapy and 7 breakthrough or relapse patients. The rapid responders included 28 poor responders and 2 breakthrough/relapsers. The rapid and slow progressor groups were statistically indistinguishable at assignment to the control arm of HALT-C for an array of clinical parameters relevant to liver disease (Table 6). The HCV open reading frame was sequenced for each patient from two time points separated by three years. Time point 1 (TP1) was nine months into the HALT-C observational period to allow HCV titers to rebound from failed interferon-based antiviral therapy that all patients received prior to randomization into the interventional or control arms of the study. The second time point (TP2) was at the end of the 45 month observational period. Consensus sequences for the HCV coding region were obtained from serum-derived RNA by direct sequencing of overlapping nested reverse transcriptase-pcr amplicons as previously described [26,69,70]. We were unable to sequence the full coding region from two rapid progressor and one slow progressor samples for time point 2, so these sequences were excluded from analyses involving time point 2. HCV positional differences associated with rate of disease progression There were 15 positions in the TP1 and 13 positions in the TP2 sequences where the distributions of amino acids were significantly different between rapid and slow progressors, with seven of those positions overlapping between time points (Table 7). To help evaluate the likelihood that these may be spurious associations, we generated five sets of paired sequence groups where the 60 sequences were randomly assigned to one of two groups, with both groups containing 30 sequences. The number of positions that were significantly different between these pairs of randomized sequence sets ranged from 14 to 25, with a mean of P-values ranged from to 0.049, with a mean of These values were very similar to the values seen when the rapid and slow progressor sequences were compared, suggesting that the differences in Table 7 are unlikely to reflect important biological variations associated with rate of disease progression. HCV consensus sequence diversity differences are not associated with rate of disease progression Pairwise genetic distances were calculated for the sequences in the rapid and slow groups for both TP1 and TP2 as we did for the cancer cohort. No significant differences were observed in the average pairwise distances between the rapid and slow progressors for either time point or between time points. Positions of variance relative to a population-wide reference were identified for rapid and slow progressors at both time points, and no significant differences were found at either time point between the two groups. This was true both when the entire polyprotein was evaluated as a single unit and when the viral genes were considered individually. The paired sequences from TP1 and TP2 for each patient were compared and the numbers of mutations at the protein level were determined for each pair. There were no significant differences in the number of mutations during the three years between TP1 and TP2 between the rapid and slow progressors. This was true when the full polyprotein, each individual gene, or just the hypervariable regions 1 and 2 in E2 were compared. Selective pressures associated with disease progression The rapid and slow progressor sequences were examined for codon selection differences using SLAC method as we did for the cancer cohort. Far more codons were under negative selection PLOS ONE 8 July 2014 Volume 9 Issue 7 e103748

11 Table 6. Baseline characteristics of patients from whom the rapid and slow progressor samples were derived. Slow progressors Rapid progressors P value 1 Number of patients Age (mean 6 SD) ns 2 Female sex (% of patients) ns Duration of exposure to HCV (yr) (mean 6 SD) ns Race or ethnic group (% of patients) 3 ns White Black Hispanic Body-mass index (BMI) (mean 6 SD) ns Diabetes (% of patients) ns Lifetime alcohol consumption (no. of drinks) (median) ns 5 Lifetime alcohol consumption (no. of drinks) (interquartile range) Baseline serum HCV RNA (log 10 IU/ml) (mean 6 SD) ns Serum alanine aminotransferase (ALT) (U/liter) (mean 6 SD) ns Ratio of the patient s alanine aminotransferase (ALT) level to the upper limit of normal (ULN) ns (mean 6 SD) Total serum bilirubin (mg/dl) (mean 6 SD) ns Serum albumin (g/dl) (mean 6 SD) ns Prothrombin time (INR) (mean 6 SD) ns Ishak fibrosis score 6 (mean 6 SD) ns Ishak inflammation score 7 (mean 6 SD) ns Esophageal varices (% of patients) ns 1 T-test or chi-square test unless otherwise indicated. 2 Non-significant (p,0.05). 3 Race or ethnic group was self-reported. 4 BMI is the weight in kilograms divided by the square of the height in meters. 5 Wilcoxon rank-sum test. 6 The Ishak fibrosis score range is 0 (no fibrosis) to 6 (cirrhosis). 7 The Ishak inflammation score range is 0 (best) to 18 (worst). doi: /journal.pone t006 (,1100 in both groups) than were under positive selection at both time points. The number of codons under positive selection was higher for slow progressors compared to rapid progressors at both time points (14 vs. 7 at TP1; 18 vs. 8 at TP2). Most of the codons under positive selection for the rapid progressors overlapped with those identified for the slow progressors (Table 8). To help evaluate whether these differences in the number of codons under selection may be related to the rate of disease progression, we randomly selected six sets of 30 subtype 1a sequences from Genbank and examined their positive selection patterns. The number of sites under positive selection for the control sets ranged from 8 to 22 with an average All but three of the codons for TP1 and seven of the codons for TP2 were under positive selection in one or more of the control sets (Table 8). Therefore, almost all of the sites under positive selection in the HALT-C dataset were not preferentially associated with the rate of disease progression. However, the slow progressor sequences were under greater positive selection pressure compared to the rapid progressor sequences. UU and UA dinucleotide frequency differences are not associated with rate of disease progression The ratio of observed/predicted dinucleotide frequency was determined for every possible dinucleotide pair for each sample as before. As with the HCC cohort, all of the samples had reduced frequencies of UA and UU dinucleotides, but there were no significant differences in the observed/expected UU or UA ratios between the rapid and slow progressors, either within a time point or when the time points were combined. Only the AU dinucleotide had a statistically significant difference in the observed/predicted ratio between rapid (0.919) and slow (0.933) progressors (p,0.001). Although this is statistically significant, the magnitude of the change is very small and hence the difference unlikely to be biologically significant. Amino acid covariance patterns are not associated with rate of disease progression Finally, we generated amino acid covariance networks for the HCV sequences from the rapid and slow progressors at both time points using the same methods used for the HCC cohort. As has been observed for other HCV sequence sets [27,29,31,75], amino acid covariance networks were identified that involved residue positions from all 10 proteins and that had a hub-and-spoke topology. For both time points, network parameters including number of nodes, number of edges, mean number of neighbors, density and clustering coefficient were very similar between rapid and slow progressors (Table 9). About half of the covarying residue pairs and over 80% of the residue positions overlapped between rapid and slow progressor networks, indicating that the networks were very similar (data not shown). The networks generated at the PLOS ONE 9 July 2014 Volume 9 Issue 7 e103748

12 Table 7. Positions where the distribution of amino acids differ between the rapid and slow progressor sequences. Time point 1 Time point 2 Position Protein P value Position Protein P value 308 E E E E E E E NS E NS p NS NS NS4B NS NS4B NS4B NS5A NS4B NS5A NS4B NS5A NS5A NS5A NS5A NS5B NS5A NS5B doi: /journal.pone t007 two time points for the rapid progressors were almost indistinguishable, as were the two networks for the slow progressors. Therefore, covariance network analysis failed to identify differences between the rapid and slow progressor sequences. Discussion HCV is genetically very diverse, and viral genetic variation is a major contributor to virulence in many viral pathogens. However, evidence for or against HCV s high genetic variation leading to differential virulence within a genotype is limited. Here, we examined HCV genetic variation in the full viral protein coding region to determine if genetic differences in HCV genotype 1 are associated with the development of HCC or the rate of disease progression. In sharp contrast to the strong associations we and others found between viral diversity and covariation patterns with response to interferon a-based therapy [25 27,30,31], very few HCV genetic associations were found with development of HCC or the rate of disease progression. HCV genetic associations with HCC The HCC and cirrhotic control sequences were very similar, but we were able to identify two differences between them. First, there were 25 positions where the distribution of amino acids in the HCC and cirrhotic sequences were significantly different, which was more than the differences observed between control sequence sets in which these sequences were randomly re-sorted without regard for disease state. Four of these positions were within the p7 gene (Table 2). This clustering of differences within the very small p7 protein (63 residues) may imply a previously undefined role for this ion channel protein in the progression to HCC within a badly diseased liver. Three studies from Japan previously examined the entire HCV ORF for positions of variability associated with HCC [64,65] [66]. These studies each identified up to nine positions in the core, E2, NS2, NS3 and NS5A genes where the amino acid distribution differed significantly between viruses from HCC patients and asymptomatic controls. The positions of skewed amino acid distributions we found were not the same as the sites found by the Japanese investigators. Together, these observations indicate that there may be some sites in the HCV genome where sequence differences are associated with HCC, but the inconsistency in the positions identified implies that it is unlikely such differences will be informative mechanistically or diagnostically. Second, we found eight positions under positive selection that were unique to either the HCC or cirrhotic control groups that were not under positive selection in randomly selected sets of genotype 1b sequences (Table 3). These positions may therefore be under evolutionary pressures associated with these advanced disease states. Nucleotide sequence variations at eleven positions within the core gene have been previously associated with HCC [54], and amino acid variations at core positions 70 and 91 are associated with HCC in HCV 1b-infected patients, especially in Japan [55 57,66]. However, sequences corresponding to the non-hcc signature strongly predominated at eight of these eleven nucleotide positions in both the cirrhotic control sequences and the HCC sequences. This observation was confirmed by evaluating an external set of cirrhotic patients [62] Table 5). The three exceptions were at nucleotides 78, 209 (within codon 70), and nucleotide 271 (in codon 91). Here, the HCC signature predominated in both the HCC and cirrhotic sequences. The previous studies that identified genetic associations in the HCV core gene with HCC used non-cirrhotic patients as controls [54,58], but 80 90% of HCV-associated HCCs develop within a cirrhotic liver [82]. The equal prevalence of the cancer-associated genetic signatures in the HCC and cirrhotic control sequences indicates that these signatures are more likely to reflect an adaptation of HCV to a cirrhotic liver rather than direct associations with HCC. We found no significant differences in the covariance networks between the HCC and cirrhotic controls. This result in is contrast to our previous covariance network analyses of HCV that identified strong signatures associated with early response to interferon-based treatment [27]. Furthermore, a different covari- PLOS ONE 10 July 2014 Volume 9 Issue 7 e103748